A lifetime reliability prediction of engine components can be enhanced by considering microstructural damage evolution. Cracks start to arise at microscopic level and propagate through the material leading to failure. A suitable approach to take microstructural properties into account is the so-called FE2-method [1,2]. The main approach of this multiscale framework is the replacement of a constitutive material model on the macroscopic level by effective properties, obtained by a homogenization process in each integration point based on a representative volume element (RVE). In order to reflect the overall properties, this RVE should reflect the main characteristics of the underlying complex microstructure by considering the morphology, pores or microcracks. Additionally, the interaction between the different phases and resulting microscopic crack paths are captured within the homogenization process. For the simulation of the cracks, we use a phase field model, where the sharp topology of a crack is approximated by a diffusive transition zone described by a phase field variable. This variable characterizes the degree of damage of a microscopic material point. In the used framework of a selective homogenization approach [3,4], the phase field variable only exists on microscopic level, which significantly reduces the computational costs. Furthermore, the still very expensive multiscale simulations can be massively accelerated by novel machine learning algorithms, which replace the simulations and homogenization processes of each RVE and make precise predictions of the corresponding effective properties [5]. A complete workflow from the construction of RVEs to the training of a machine learning model and the application to components will be presented. REFERENCES [1] V. Kouznetsova, W.A.M. Brekelmans and F.P.T. Baaijens. An approach to micro-macro modeling of heterogeneous materials, Computational Mechanics, Vol. 27, pp. 37-48, 2001. [2] J. Schröder, M. Labusch and M.-A. Keip. Algorithmic Two-Scale Transition for Magneto-Electro-Mechanically Coupled Problems, FE2-Scheme: Localization and Homogenization, Computer Methods in Applied Mechanics and Engineering, Vol. 302, pp. 253-280, 2016. [3] M. Labusch, L. Reischmann, M. Meurer and S. Reh. A Multiscale Approach for Prediction of Failure Probabilities of Engine Components, Journal of Engineering for Gas Turbines and Power, Vol. 148 (4), pp. 041021, 2026. [4] R. Bharali, F. Larsson and R. Jänicke. Computational H